Re: [External] Re: [PATCH 1/2] ext4: prevent used blocks from being allocated during fast commit replay

[harshad shirwadkar]

Ah, I see okay. Yeah, that makes sense. We do need to update
fc_regions_valid for every addition during the replay phase. I think
it may be helpful to add some comments describing this behavior in the
code. But other than that, I think what you're doing is fine.

Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@xxxxxxxxx>

On Fri, Jan 7, 2022 at 6:09 PM Xin Yin <yinxin.x@xxxxxxxxxxxxx> wrote:
>
> On Sat, Jan 8, 2022 at 4:26 AM harshad shirwadkar
> <harshadshirwadkar@xxxxxxxxx> wrote:
> >
> > On Wed, Jan 5, 2022 at 6:45 PM Xin Yin <yinxin.x@xxxxxxxxxxxxx> wrote:
> > >
> > > during fast commit replay procedure, we clear inode blocks bitmap in
> > > ext4_ext_clear_bb(), this may cause ext4_mb_new_blocks_simple() allocate
> > > blocks still in use.
> > >
> > > make ext4_fc_record_regions() also record physical disk regions used by
> > > inodes during replay procedure. Then ext4_mb_new_blocks_simple() can
> > > excludes these blocks in use.
> > >
> > > Signed-off-by: Xin Yin <yinxin.x@xxxxxxxxxxxxx>
> > > ---
> > >  fs/ext4/ext4.h        |  2 ++
> > >  fs/ext4/extents.c     |  4 ++++
> > >  fs/ext4/fast_commit.c | 11 ++++++++---
> > >  3 files changed, 14 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
> > > index 82fa51d6f145..7b0686758691 100644
> > > --- a/fs/ext4/ext4.h
> > > +++ b/fs/ext4/ext4.h
> > > @@ -2932,6 +2932,8 @@ bool ext4_fc_replay_check_excluded(struct super_block *sb, ext4_fsblk_t block);
> > >  void ext4_fc_replay_cleanup(struct super_block *sb);
> > >  int ext4_fc_commit(journal_t *journal, tid_t commit_tid);
> > >  int __init ext4_fc_init_dentry_cache(void);
> > > +int ext4_fc_record_regions(struct super_block *sb, int ino,
> > > +                    ext4_lblk_t lblk, ext4_fsblk_t pblk, int len, int replay);
> > >
> > >  /* mballoc.c */
> > >  extern const struct seq_operations ext4_mb_seq_groups_ops;
> > > diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
> > > index c3e76a5de661..9b6c76629c93 100644
> > > --- a/fs/ext4/extents.c
> > > +++ b/fs/ext4/extents.c
> > > @@ -6096,11 +6096,15 @@ int ext4_ext_clear_bb(struct inode *inode)
> > >
> > >                                         ext4_mb_mark_bb(inode->i_sb,
> > >                                                         path[j].p_block, 1, 0);
> > > +                                       ext4_fc_record_regions(inode->i_sb, inode->i_ino,
> > > +                                                       0, path[j].p_block, 1, 1);
> > >                                 }
> > >                                 ext4_ext_drop_refs(path);
> > >                                 kfree(path);
> > >                         }
> > >                         ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
> > > +                       ext4_fc_record_regions(inode->i_sb, inode->i_ino,
> > > +                                       map.m_lblk, map.m_pblk, map.m_len, 1);
> > >                 }
> > >                 cur = cur + map.m_len;
> > >         }
> > > diff --git a/fs/ext4/fast_commit.c b/fs/ext4/fast_commit.c
> > > index 23d13983a281..f0cd20f5fe5e 100644
> > > --- a/fs/ext4/fast_commit.c
> > > +++ b/fs/ext4/fast_commit.c
> > > @@ -1567,13 +1567,15 @@ static int ext4_fc_replay_create(struct super_block *sb, struct ext4_fc_tl *tl,
> > >   * Record physical disk regions which are in use as per fast commit area. Our
> > >   * simple replay phase allocator excludes these regions from allocation.
> > >   */
> > > -static int ext4_fc_record_regions(struct super_block *sb, int ino,
> > > -               ext4_lblk_t lblk, ext4_fsblk_t pblk, int len)
> > > +int ext4_fc_record_regions(struct super_block *sb, int ino,
> > > +               ext4_lblk_t lblk, ext4_fsblk_t pblk, int len, int replay)
> > Can you explain a bit why this replay parameter is needed here? This
> > function simply reallocs the regions array if it doesn't have enough
> > space. I am not sure why we need to change that behavior.
>
> ext4_fc_record_regions() originally only used during scan phase,  and
> set fc_regions_valid = fc_regions_use when getting a TAIL tag. Now we
> also use it during the replay phase,   and need to update
> fc_regions_valid in this case, because ext4_fc_replay_check_excluded()
> uses fc_regions_valid for regions checking.
> Please correct me if I'm wrong.
>
> > >  {
> > >         struct ext4_fc_replay_state *state;
> > >         struct ext4_fc_alloc_region *region;
> > >
> > >         state = &EXT4_SB(sb)->s_fc_replay_state;
> > > +       if (replay && state->fc_regions_used != state->fc_regions_valid)
> > > +               state->fc_regions_used = state->fc_regions_valid;
> > >         if (state->fc_regions_used == state->fc_regions_size) {
> > >                 state->fc_regions_size +=
> > >                         EXT4_FC_REPLAY_REALLOC_INCREMENT;
> > > @@ -1591,6 +1593,9 @@ static int ext4_fc_record_regions(struct super_block *sb, int ino,
> > >         region->pblk = pblk;
> > >         region->len = len;
> > >
> > > +       if (replay)
> > > +               state->fc_regions_valid++;
> > > +
> > >         return 0;
> > >  }
> > >
> > > @@ -1938,7 +1943,7 @@ static int ext4_fc_replay_scan(journal_t *journal,
> > >                         ret = ext4_fc_record_regions(sb,
> > >                                 le32_to_cpu(ext.fc_ino),
> > >                                 le32_to_cpu(ex->ee_block), ext4_ext_pblock(ex),
> > > -                               ext4_ext_get_actual_len(ex));
> > > +                               ext4_ext_get_actual_len(ex), 0);
> > >                         if (ret < 0)
> > >                                 break;
> > >                         ret = JBD2_FC_REPLAY_CONTINUE;
> > > --
> > > 2.20.1
> > >